Hi all, I investigated more in detail the efficiency dependence on purity I mentioned in http://www.slac.stanford.edu/cgi-bin/lwgate/VUB-RECOIL/archives/vub-recoil.200208/Author/article-80.html In http://www.slac.stanford.edu/BFROOT/www/Physics/Analysis/AWG/InclusiveSL/islrecoil/QA_09032002.html you can find the results for 70fb-1 as far as breco yields, optimization and efficiency stuff is concerned. I would like two show you two sets of plots. The last column of the table called "Ratio's for Data" shows the ratio Nsignal events (after allcuts) / Nsignal events (after lepton cuts) in bins of the purity of the single mode, that is actually a measurement of the efficiency vs the purity of our sample. If there is a problem and a lower efficiency, it is present only in the last 2-3 bins (15%<pur<10%, 10%<pur<5%, 5%<pur). This problem can be due either to Physics (efficiency depends for some reasons to the reconstructed mode) or to a mes fit problem. In order to check the first possibility, I checked the same distribution on generic MC. In the last column of "Ratio's for Generic MC" table you can find the same plots for the generic MC (unfortunately the bins do not containt exactly the same modes as the data ones). As shown, the dependency seems to be gone. Then the first possibility could be excluded. As a consequence, this effect could be only an artefact of the mes fits (and this is the impression you have if you look at the single mes fit) and I checked that is due to the combinatoric shape parametrization (fixing the signal shape I got the same dependece), that is modeled using an argus function. I will perform more studies as far as the fit parametrization is concerned but I have to say that the problem we are observing is not a big problem since the purity optimization I performed few days ago basically removes the last two bins of the efficiency plots. Daniele